Electronic dance music genre classification
|Title||Electronic dance music genre classification|
|Publication Type||Master Thesis|
|Year of Publication||2008|
The purpose of this thesis is to contribute to the automatic classification of Electronic Dance Music genres, and deal with the existing feature extractors and classifiers to finally use existing algorithms. The state of the art on music classification and the overviews of algorithms and features used for current approach are presented. The 250 excerpts were taken from a Master Thesis presented on 2007 by a Priit Kirss at the University of Jyväskylä which provides with five genres: Deep House, Techno, Uplifting Trance, Drum and Bass and Ambient. We want to compare our feature extractors and how they deal with these genres and what features became more useful, so our approach consists of two main steps: deeply and objectively describe these genres and available features, and evaluate how they contribute to the classification.
The experiments carried out testing different combinations of features provided by MTG and MIR Toolbox. Also we have tested a higher level statistical rhythm feature set (Haro, 2008), which showed very promising results In this work we have tested the relevance of features among the objective definition of the genres. From the experiments, we have tested that the MFCCs have been proven to add noise to the classification, while the rhythm features increment the accuracy.
Training the MTG feature set with SVM algorithm a 90.68% was reached.